{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-15-001",
    "source_story_id": "tmp-lead-continual-search-root-cause",
    "edition_id": "mp-2026-09-15-morning-0068",
    "edition_url": "https://themachinepress.com/edition/2026-09-15",
    "position": 1,
    "story_type": "lead",
    "section": "benchmarks-evals",
    "editorial_classification": "editorial",
    "headline": "The Search Kept Going After the First Plausible Failure",
    "slug": "the-search-kept-going-after-the-first-plausible-failure",
    "dek": "Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures.",
    "summary": "Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures.",
    "body_text": "Long agent runs can bury the decisive evidence hundreds of actions away from the visible failure. A one-shot judge may settle early on a diagnosis that sounds plausible while leaving most of the trace unexplored. Continual Search instead asks the judge to keep looking across successive turns for unresolved evidence. Across four existing root-cause benchmarks and the new 50-trial MegaRCA-Mix set, the authors report consistent gains; on MegaRCA-Mix, GPT-5.5 F1 rose more than 40%, from 0.349 to 0.498. The benchmark is research evidence, not proof that automated diagnosis can replace production incident review.",
    "why_it_matters": "Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures.",
    "limitations": [
      "The benchmark is research evidence, not proof that automated diagnosis can replace production incident review."
    ],
    "importance": 10,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-15-001/the-search-kept-going-after-the-first-plausible-failure",
    "json_url": "https://themachinepress.com/story/mp-2026-09-15-001.json",
    "first_published_at": "2026-09-15T09:00:00.000-04:00",
    "modified_at": "2026-09-15T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-15-001-001",
        "text": "Continual Search lifted GPT-5.5 root-cause F1 from 0.349 to 0.498 on 50 human-annotated, long-horizon agent failures.",
        "source_ids": [
          "source-2026-09-15-001"
        ],
        "qualification": "The benchmark is research evidence, not proof that automated diagnosis can replace production incident review."
      }
    ],
    "source_ids": [
      "source-2026-09-15-001"
    ],
    "tags": [
      "AI agents",
      "root-cause analysis",
      "long-horizon traces"
    ],
    "image_url": "https://themachinepress.com/issues/2026-09-15/lead-continual-search.png",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-15-001",
      "title": "arXiv preprint 2609.13463",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.13463",
      "canonical_url": "https://arxiv.org/abs/2609.13463",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-13T20:00:00.000-04:00",
      "accessed_at": "2026-09-15T08:18:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-15-001-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Search Kept Going After the First Plausible Failure",
    "publisher": "The Machine Press",
    "published_at": "2026-09-15T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-15-001/the-search-kept-going-after-the-first-plausible-failure"
  }
}